Image Enhancement in Reconstructed Photoacoustic Microscopy Images Using Deep Learning

Abstract

Photo-acoustic Microscopy (PAM) is an imaging technique that is used in medical imaging. It is centred on the effects in which the sound waves follow the light absorption in an object, referred to as the photo-acoustic effect. In various other imaging modalities, image reconstruction is a strenuous task when compared with photo-acoustic microscopy images. Image reconstruction is simple in the case of PAM images. For medical imaging, various artifacts and noise may arise. It is crucial to perform image reconstruction before retrieving the original images. The prime technical challenge associated with photo- acoustic microscopy images (PAM) is the imperative requital between spatial definition and visualization speed. In this paper, researcher suggest a novel implementation of deep learning concept to perform image enhancement in reconstructed PAM images. Aiming to perform the task, researcher choose the Fully Dense U-net (FD U-net) model, which gives the best result, and also used wave-let based image de-noising technique and median filtering, which is one of the most powerful tools for image enhancement. To perform so, we need certain under-sampled images, so to acquire the under-sampled images, researcher synthetically down-sampled the fully-sampled PAM images in order to perform image reconstruction. It supports us in performing both training and testing our model under various imaging constraints. It improves the quality of the images, so it can be used for medical diagnosis. Another important aspect of the wavelet-based denoising techniques is their lower computational complexity. Researcher proposed a novel statistic-based Blind/Reference-less Image Spatial Quality Evaluator (BRISQUE), which is an image quality assessment method. Thus, by using BRISQUE, the quality of the image can be determined. The overall results and analysis collectively show the sturdy performance of our proposed work to reconstruct PAM images. By implementing the model, it can effectively reduce the imaging time and also can improve the quality of the images.